Spaces:
Sleeping
Sleeping
Create oh_yeah.py
Browse files- lab/oh_yeah.py +121 -0
lab/oh_yeah.py
ADDED
@@ -0,0 +1,121 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import pandas as pd
|
3 |
+
import os
|
4 |
+
from dotenv import load_dotenv
|
5 |
+
from llama_index.core import Settings, VectorStoreIndex, SimpleDirectoryReader
|
6 |
+
from llama_index.core.readers.base import BaseReader
|
7 |
+
from llama_index.readers.file.paged_csv.base import PagedCSVReader
|
8 |
+
from llama_index.embeddings.openai import OpenAIEmbedding
|
9 |
+
from llama_index.llms.openai import OpenAI
|
10 |
+
from llama_index.vector_stores.faiss import FaissVectorStore
|
11 |
+
from llama_index.core.ingestion import IngestionPipeline
|
12 |
+
from langchain_community.document_loaders.csv_loader import CSVLoader
|
13 |
+
from langchain_community.vectorstores import FAISS as LangChainFAISS
|
14 |
+
from langchain.chains import create_retrieval_chain
|
15 |
+
from langchain.chains.combine_documents import create_stuff_documents_chain
|
16 |
+
from langchain_core.prompts import ChatPromptTemplate
|
17 |
+
from langchain_openai import OpenAIEmbeddings, ChatOpenAI
|
18 |
+
import faiss
|
19 |
+
|
20 |
+
# Load environment variables
|
21 |
+
os.environ["OPENAI_API_KEY"] = os.getenv("OPENAI_API_KEY")
|
22 |
+
|
23 |
+
# Global settings for LlamaIndex
|
24 |
+
EMBED_DIMENSION = 512
|
25 |
+
Settings.llm = OpenAI(model="gpt-3.5-turbo")
|
26 |
+
Settings.embed_model = OpenAIEmbedding(model="text-embedding-3-small", dimensions=EMBED_DIMENSION)
|
27 |
+
|
28 |
+
# Streamlit app
|
29 |
+
st.title("Chat w CSV Files - LangChain Vs LlamaIndex ")
|
30 |
+
|
31 |
+
# File uploader
|
32 |
+
uploaded_file = st.file_uploader("Upload a CSV file", type=["csv"])
|
33 |
+
if uploaded_file:
|
34 |
+
# Save the uploaded file temporarily
|
35 |
+
temp_file_path = f"temp_{uploaded_file.name}"
|
36 |
+
with open(temp_file_path, "wb") as temp_file:
|
37 |
+
temp_file.write(uploaded_file.getbuffer())
|
38 |
+
|
39 |
+
# Read and preview CSV data
|
40 |
+
data = pd.read_csv(temp_file_path)
|
41 |
+
st.write("Preview of uploaded data:")
|
42 |
+
st.dataframe(data)
|
43 |
+
|
44 |
+
# Tabs
|
45 |
+
tab1, tab2 = st.tabs(["Chat w CSV using LangChain", "Chat w CSV using LlamaIndex"])
|
46 |
+
|
47 |
+
# LangChain Tab
|
48 |
+
with tab1:
|
49 |
+
st.subheader("LangChain Query")
|
50 |
+
loader = CSVLoader(file_path=temp_file_path)
|
51 |
+
docs = loader.load_and_split()
|
52 |
+
|
53 |
+
# Preview the first document
|
54 |
+
if docs:
|
55 |
+
st.write("Preview of a document chunk (LangChain):")
|
56 |
+
st.text(docs[0].page_content)
|
57 |
+
|
58 |
+
# LangChain FAISS VectorStore
|
59 |
+
langchain_index = faiss.IndexFlatL2(EMBED_DIMENSION)
|
60 |
+
langchain_vector_store = LangChainFAISS(
|
61 |
+
embedding_function=OpenAIEmbeddings(),
|
62 |
+
index=langchain_index,
|
63 |
+
)
|
64 |
+
langchain_vector_store.add_documents(docs)
|
65 |
+
|
66 |
+
# LangChain Retrieval Chain
|
67 |
+
retriever = langchain_vector_store.as_retriever()
|
68 |
+
system_prompt = (
|
69 |
+
"You are an assistant for question-answering tasks. "
|
70 |
+
"Use the following pieces of retrieved context to answer "
|
71 |
+
"the question. If you don't know the answer, say that you "
|
72 |
+
"don't know. Use three sentences maximum and keep the "
|
73 |
+
"answer concise.\n\n{context}"
|
74 |
+
)
|
75 |
+
prompt = ChatPromptTemplate.from_messages(
|
76 |
+
[("system", system_prompt), ("human", "{input}")]
|
77 |
+
)
|
78 |
+
question_answer_chain = create_stuff_documents_chain(ChatOpenAI(), prompt)
|
79 |
+
langchain_rag_chain = create_retrieval_chain(retriever, question_answer_chain)
|
80 |
+
|
81 |
+
# Query input for LangChain
|
82 |
+
query = st.text_input("Ask a question about your data (LangChain):")
|
83 |
+
if query:
|
84 |
+
answer = langchain_rag_chain.invoke({"input": query})
|
85 |
+
st.write(f"Answer: {answer['answer']}")
|
86 |
+
|
87 |
+
# LlamaIndex Tab
|
88 |
+
with tab2:
|
89 |
+
st.subheader("LlamaIndex Query")
|
90 |
+
csv_reader = PagedCSVReader()
|
91 |
+
reader = SimpleDirectoryReader(
|
92 |
+
input_files=[temp_file_path],
|
93 |
+
file_extractor={".csv": csv_reader},
|
94 |
+
)
|
95 |
+
docs = reader.load_data()
|
96 |
+
|
97 |
+
# Preview the first document
|
98 |
+
if docs:
|
99 |
+
st.write("Preview of a document chunk (LlamaIndex):")
|
100 |
+
st.text(docs[0].text)
|
101 |
+
|
102 |
+
# Initialize FAISS Vector Store
|
103 |
+
llama_faiss_index = faiss.IndexFlatL2(EMBED_DIMENSION)
|
104 |
+
llama_vector_store = FaissVectorStore(faiss_index=llama_faiss_index)
|
105 |
+
|
106 |
+
# Create the ingestion pipeline and process the data
|
107 |
+
pipeline = IngestionPipeline(vector_store=llama_vector_store, documents=docs)
|
108 |
+
nodes = pipeline.run()
|
109 |
+
|
110 |
+
# Create a query engine
|
111 |
+
llama_index = VectorStoreIndex(nodes)
|
112 |
+
query_engine = llama_index.as_query_engine(similarity_top_k=3)
|
113 |
+
|
114 |
+
# Query input for LlamaIndex
|
115 |
+
query = st.text_input("Ask a question about your data (LlamaIndex):")
|
116 |
+
if query:
|
117 |
+
response = query_engine.query(query)
|
118 |
+
st.write(f"Answer: {response.response}")
|
119 |
+
|
120 |
+
# Cleanup temporary file
|
121 |
+
os.remove(temp_file_path)
|